The case for macroeconometric models
Macroeconometric models, in many ways the flagships of the economics profession in the 1960s, came under increasing attack from both theoretical economics and practitioners in the late 1970s. The onslaught came on a wide front: lack of microeconomic theoretical foundations, ad hoc modelling of expectations, lack of identification, neglect of dynamics and non-stationarity, and poor forecasting properties. As a result, by the start of the 1990s, the status of macroeconometric models had declined markedly, and had fallen completely out of (and with!) academic economics. Specifically, it has become increasingly rare that university programmes in economics give courses in large-scale empirical macroeconomic modelling.
Nevertheless, unlike the dinosaurs which they often have been likened to, macroeconometric models never completely disappeared from the scene. Moreover, if we use the term econometric model in a broad sense, it is fair to say that such models continue to play a role in economic policy. Model building and maintenance, and model based economic analyses, continue to be an important
part of many economists’ working week, either as a producer (e. g. member of modelling staff) or as a consumer (e. g. chief economists and consultants). Thus, the discipline of macroeconometric modelling has been able to adapt to changing demands, both with regards to what kind of problems users expect that models can help them answer, and with regard to quality and reliability.
Consider, for example, the evolution of Norwegian macroeconometric models (parallel developments no doubt have taken place in other countries): the models of the 1960s were designed to meet the demands of governments which attempted to run the economy through regulated markets. Today’s models have adapted to a situation with liberalised financial and credit markets. In fact, the process of deregulation has resulted in an increased demand for econometric analysis and forecasting.
The recent change in monetary policy towards inflation targeting provides an example of how political and institutional changes might affect econometric modelling. The origins of inflation targeting seem to be found in the practical and operational issues which the governments of small open economies found themselves with after installing floating exchange rate regimes. As an alternative to the targeting of monetary aggregates, several countries (New Zealand, Canada, United Kingdom, and Sweden were first) opted for inflation targeting, using the interest rate as the policy instrument. In the literature which followed in the wake of the change in central bank practice (see, for example, Svensson 2000), it was made clear that under inflation targeting, the central bank’s conditional inflation forecast becomes the operational target of monetary policy. At the back of the whole idea of inflation targeting is therefore the assumption that the inflation forecast is significantly affected by adjustment of the interest rate ‘today’. It follows that the monetary authority’s inflation forecasts have to be rooted in a model (explicit or not) of the transmission mechanism between the interest rate and inflation.
This characterisation of inflation targeting leads to a set of interesting questions, around which a lively debate evolves. For example: how should the size and structure of the model be decided, and its parameters quantified, that is, by theoretical design, by estimation using historical data or by some method of calibration—or perhaps by emulating the views of the ‘monetary policy committee’ (since at the end of the day the beliefs of the policy makers matter). A second set of issues follows from having the forecasted rate of inflation (rather than the current or historical rate) as the target. As emphasised by, for example, Clements and Hendry (19956), modelling and forecasting are distinct processes (see also Chapter 11). In particular non-stationarities which are not removed by differencing or cointegration impinge on macroeconomic data. One consequence is that even well-specified policy models produce intermittent forecast failure, by which we in this book mean a significant deterioration in forecast quality relative to within sample tracking performance (see Clements and Hendry 19996: ch. 2). Both theory and practical experience tell us that the source of forecast failure is usually to be found in shifts in the means of equilibrium relationships and in the growth rates of exogenous variables. Neither of these factors affect a model’s usefulness in policy analysis, yet either of them can destroy the model’s forecasts, unless the model user is able to correct them (e. g. by intercept corrections).
The integration of modelling, policy analysis, and forecasting in the mandate given to an inflation targeting central bank raises some important issues. For example, it must be decided to what extent the policy model should affect the forecasts, and how forecasts are best robustified in order to reduce the hazards of forecast-based interest rate setting.
Inflation targeting has already spurred a debate about the role of econometric specification and evaluation of models—that is, not only as an aid in the preparation of inflation forecasts, but also as a way of testing, quantifying, and elucidating the importance of transmission mechanisms in the inflationary process. In this way, inflation targeting actually moves the discussion about the quality and usefulness of econometric methodology and practice into the limelight of the economic policy debate (see Bardsen et al. 2003).
However, even though a continued and even increasing demand for macroeconometric analysis is encouraging for the activity of macroeconometric modelling, it cannot survive as a discipline within economics unless the models reflect the developments in academic research and teaching. But, also in this respect macroeconometric modelling has fared much better than many of its critics seem to acknowledge. Already by the end of the 1980s, European macroeconometric models had a much better representation of price - and wagesetting (i. e. the supply-side) than before. There was also marked improvement in the modelling of the transmission mechanism between the real and financial sectors of the economy (see, for example, Wallis 1989). In the course of the last 20 years of the last century macroeconometric models also took advantage of the methodological and conceptual advances within time-series econometrics. Use of dynamic behavioural equations are now the rule rather than the exception. Extensive testing of mis-specification is usually performed. The dangers of spurious regressions (see Granger and Newbold 1974) have been reduced as a consequence of the adoption of new inference procedures for integrated variables. No doubt, an important factor behind these advances has been the development of (often research based) software packages for estimation, model evaluation, and simulation.
In an insightful paper about the trends and problems facing econometric models, the Norwegian economist Leif Johansen stated that the trendlike development in the direction of more widespread use of econometric models will hardly be reversed completely (see Johansen 1982). But Johansen also noted that both the models’ own conspicuous failures from time to time, and certain political developments, will inflict breaks or temporary setbacks in the trend. However, we think that we are in line with Johansen’s views when we suggest that a close interchange between academic economics, theoretical econometrics, and software development are key elements that are necessary to sustain macroeconomic modelling. The present volume is meant as a contribution to macroeconomic modelling along these lines.
Four themes in particular are emphasised in this book:
(1) methodological issues of macroeconometric models;
(2) the supply-side of macroeconometric models;
(3) the transmission mechanism;
(4) the forecasting properties of macroeconometric models.
In the following, we review the main issues connected to these themes, and explain where they are covered in the book.